This paper presents a novel technique for controlling the current of Switched Reluctance Motor (SRM) drives based on reinforcement learning. The proposed current controller is based on a new scheduled Q-learning. Solving the infinite horizon linear quadratic tracker (LQT) problem for an unknown dynamic system of SRM drive, a new control scheme relying on the Q-learning algorithm is introduced for that purpose. The reference current generator of the SRM drive has been incorporated into the augmented system. A Q-learning algorithm is implemented to obtain the optimum solution of Algebraic Riccati Equation (ARE) with the absence of any data about system dynamics of SRM or the reference current generator. Additionally, a scheduling mechanism switches between Q matrices to allow for a nonlinear control using a table of Q-learning cores. After the introduction of the control scheme, a simulation has been designed to evaluate the performance of the proposed controller.
- Controllers,
- Digital storage,
- Electric current control,
- Electric drives,
- Electric machine control,
- Learning algorithms,
- Learning systems,
- Reinforcement learning,
- Riccati equations,
- Scheduling, Algebraic Riccati equations,
- Current controller,
- Linear quadratic trackers,
- Non linear control,
- Q-learning algorithms,
- Reference current generators,
- Scheduling mechanism,
- Switched reluctance motor drives, Reluctance motors
Available at: http://works.bepress.com/mehdi-ferdowsi/191/